Re: replacement getaddrinfo not working properly

[Matthew Woehlke]

Andrew J. Schorr wrote:
> It seems that adding AC_USE_SYSTEM_EXTENSIONS to configure.ac
> will always cause _POSIX_PTHREAD_SEMANTICS and _TANDEM_SOURCE to
> be defined (at least that is the case for autoconf version 2.61 on my
> linux FC6 host).  From /usr/share/autoconf/autoconf/specific.m4:
> [snip]
> As you can see, this defines _POSIX_PTHREAD_SEMANTICS and _TANDEM_SOURCE
> unconditionally.
>
> As a result, this logic in io.c to decide whether to include
> missing_d/getaddrinfo.h is not working as intended:
>
>    #ifdef _TANDEM_SOURCE
>    #include <arpa/inet.h>
>    #include <netdb.h>
>    #else
>    #ifdef HAVE_GETADDRINFO
>    #include <netdb.h>
>    #else
>    #include "missing_d/getaddrinfo.h"
>    #endif
>    #endif
>
> Because _TANDEM_SOURCE is defined, there is never any chance
> of including missing_d/getaddrinfo.h.
>
> This causes compilation errors on systems that do not supply getaddrinfo.
>
> I'm not certain of the best fix.  Why is it necessary to test
> for _TANDEM_SOURCE at all in this spot?  It seems that the only
> difference is whether to include <arpa/inet.h>.  Couldn't configure.ac
> just test whether this header file exists (and similarly for netdb.h),
> and just include them unconditionally if they exist (without caring
> about _TANDEM_SOURCE)?

What platform are you on? _TANDEM_SOURCE is needed for NSK targets (at 
least, for the box I have), but I doubt it should be defined for other 
targets. __TANDEM is defined by the compiler by default and could be 
used to check if the target needs _TANDEM_SOURCE. Or it could key off 
the target triple, in this case '*-tandem-*'.

--
Matthew
<insert bad pun... on second thought, better not>